Sparks Brain Preservation

The current estimates for costs to upload people to computers are very grim.
The jist is that neuromophic computing is in its research infancy and Jensen Huang and Lisa Su are not going to save anyone but their shareholders from 2003 with GPUs.
Although GPUs may be capable of creating forms of general and superintelligence: they are not the economical architecture to run computer programs emulating a biological brain.
At the brain preservation conference, I learned that Allen institute is taking one hour to simulate one second of a mouse brain on a supercomputer. This is a big problem both for iteration speed, and because if people are gonna be revived then ideally it would be realtime or faster!

Meanwhile, I've been working on figuring out how to reduce costs for my robot fleet's electronics and came accross this chip called RP2040. It's used for robots as a replacement for last gen ATmega328P microcontrollers.

That's when I had the idea.
Based on my analysis of the current costs of uploading a frog and human, it is currently not looking great for existing neuromorphic chips.
My idea is that you can use a bunch of these cheap RP2040 microcontrollers ($1 each) designed for realtime robot motor control, say 16 of them in a 4x4 grid ($16), and put them all on a PCB with APS6404L ($2 each) in a torus mesh network (total is $48 now). Working with me, Gemini has estimated that one RP2040/APS6404L pair can support simulating 50,000+ spiking neurons, and deductively that only only 20 of these tiles are necessary to simulate a frog brain in realtime. That's $48 * 20 + the JLCPCB costs of assembly and other board components. Maybe less than $1500!

Scale up and the realtime mouse brain driving a tiny car is probably in reach on a shoebox of Washingtons budget.

Since I first posted this, I've been sidequesting solving the neuromorphic computer bottleneck.
I read this overview paper.

Below, you can see an example of the structure I designed for my first project, which I am calling Frog Emissary, it's based on Grimm's The Frog Prince fairy tale.

The idea here is to design a neuromorphic computer which is modular and cheap enough to buy piecemeal on JLCPCB. This will reduce the replication costs substantially compared to various academic and enterprise long-running research projects which are not economical to replicate and get into pizza-stained 6th grader hands.



You can see in my initial conception, there are two hinges on either side of a 'Tadpole' neuromodule which can connect to adjacent Tadpoles to truly become a 'metamorphic' scalable design.

The INCC paper deadline is nearing.
I will try to fly to the United Kingdom next year and keep everyone updated on the progress on neuromorphic computers and get that integrated with work on preservation as I can.

Here's a better render of my in-progress design showing the dual-sided ring to sphere concept

Thanks for posting about your progress here. Some people believe that neuromorphic computing may be very important in the future. Myself, I don't have any expertise to say either way. It is very interesting! Best of luck with your endeavors.